The protocol of this study was approved by the Medical Review Boards of Teikyo University (approval no. 14-019-6), Tokyo Metropolitan Institute for Geriatrics and Gerontology (approval no. 6179), and Kanaji Thyroid Hospital (approval no. 9). All methods were carried out in accordance with the relevant guidelines and regulations (Declaration of Helsinki). Serum from patients with FTC (10 patients), FTA (14 patients), and follicular tumor of uncertain malignant potential (FT-UMP) (1 patient) was collected between May 2015 and January 2018, and the serum samples were stored at −80°C until use. Patient demographics are summarized in Table SI. Pathological diagnoses were made according to the WHO classification (5th edition) (9).

EVs were purified from human serum by the phosphatidylserine affinity method (10,11). Briefly, the extracellular domain of mouse Tim4 and the Fc region of human IgG1 (FUJIFILM Wako Pure Chemical, Osaka, Japan) were biotinylated and conjugated to streptavidin magnetic beads (Thermo Fisher Scientific, Waltham, MA, USA). Serum (500 µl) was centrifuged at 10,000 × g for 30 min and then filtered through a 0.22-µm filter. The filtered sample was diluted twofold with Tris-buffered saline (TBS) and incubated for 1 h at room temperature (RT) with Tim4-conjugated beads in the presence of 2 mM CaCl₂. After washing with 0.01% Tween-TBS containing 2 mM CaCl₂, the captured EVs were eluted with 2 mM EDTA in phosphate-buffered saline (PBS). Purified EVs were stored at −80°C until later use.

The protein concentrations of purified EVs were quantified using the CBQCA protein quantitation kit (Thermo Fisher Scientific) according to the manufacturer's protocol. The fluorescence intensity was measured using an EnVision plate reader (PerkinElmer, Waltham, MA, USA).

The size of the purified EVs was measured using the qNano particle analyzer (Izon Science, Christchurch, New Zealand). Purified EVs were diluted in 0.01% Tween-PBS buffer and measured using an NP100 nanopore. Calibration was performed using CPC100 as a standard control. Data were analyzed using Izon control software (version 3.4.2.51).

Purified EVs were mixed with sample buffer (62.5 mM Tris-HCl pH 6.8, 2% SDS, 10% glycerol, 0.01% bromophenol blue, and 5% 2-mercaptoethanol) and 130 ng of protein was loaded for SDS-polyacrylamide gel electrophoresis. Equal protein loading was used as there is no established loading control for EVs, particularly for human serum-derived EVs where the amount of purified EVs varies significantly between individuals even from the same volume. The separated proteins were then transferred to PVDF membranes (Millipore, Burlington, MA, USA). After blocking with 1% skim milk, the membranes were probed with a primary antibody, followed by a horseradish peroxidase (HRP)-linked secondary antibody. After washing, bound proteins were visualized using the ImmunoStar LD (FUJIFILM Wako Pure Chemical) as an HRP substrate and the FUSION Chemiluminescence Imaging System (Vilber, Marne-la-Vallée, France) as a detection instrument. The antibodies used in this study were as follows: anti-Alix (ab186429; Abcam, Cambridge, UK), anti-flotillin-2 (610383; BD Transduction Laboratories, Franklin Lakes, NJ, USA), anti-CD9 (13174; Cell Signaling Technology, Danvers, MA, USA), anti-RAB21 (NBP2-93771; Novus Biologicals, Centennial, CO, USA), and anti-GAPDH (2118; Cell Signaling Technology), along with secondary anti-mouse IgG (715-035-151) and anti-rabbit IgG (711-035-152) antibodies (Jackson ImmunoResearch, West Grove, PA, USA). The signal intensities of the obtained data were measured using ImageJ software (version 1.54d) (https://imagej.net/software/imagej/). Uncropped western blot images are provided in Fig. S1.

Protein digestion with trypsin was performed prior to proteomic analysis. EVs were dissolved in 0.1% Rapigest SF (Waters, Milford, MA, USA), reduced with 10 mM dithiothreitol in 50 mM triethylammonium bicarbonate (TEAB) at RT for 30 min, and then alkylated with 55 mM iodoacetamide in 50 mM TEAB at RT for 30 min in the dark. The samples were digested with trypsin (Promega, Madison, WI, USA) in 50 mM TEAB at 37°C for 16 h. The cleaved peptides were desalted using a GL-Tip SDB column (GL Sciences, Tokyo, Japan) and evaporated in a SpeedVac vacuum concentrator (Thermo Fisher Scientific). The peptides were reconstituted in 0.1% formic acid (FA) and subjected to nanoLC-MS/MS analysis.

Measurements were performed by nanoLC-MS/MS using an Ultimate 3000 RSLCnano system (Thermo Fisher Scientific) coupled to a Q Exactive hybrid quadrupole-Orbitrap mass spectrometer with a nanoESI source (Thermo Fisher Scientific), as previously described (12). The nanoLC system was equipped with a trap column (C18 PepMap 100, 0.3×5 mm, 5 µm particle size; Thermo Fisher Scientific) and an analytical column (NTCC-360/75-3-125, 75 µm × 12 cm, 3 µm particle size, C18; Nikkyo Technos, Tokyo, Japan). Peptides were separated over a period of 90 min using a gradient of water containing 0.1% FA (mobile phase A) and acetonitrile containing 0.1% FA (mobile phase B) at a flow rate of 300 nl/min. The elution gradient was set as follows: 0–3 min, 2% B; 3–93 min, 2–40% B; 93–95 min, 40–95% B; 95–105 min, 95% B; 105–107 min, 95–2% B; and 107–120 min, 2% B. The mass spectrometer was operated in data-dependent acquisition mode. All data were analyzed for protein identification and label-free quantification using Proteome Discoverer 2.4 software (Thermo Fisher Scientific). The analytical parameters used for the database search were as follows: search engine, Sequest HT; protein database, Swiss-prot (Homo Sapiens); enzyme name, trypsin (full); dynamic modification, oxidation (methionine); static modification, carbamidomethyl (cysteine); precursor mass tolerance: 10 ppm, fragment mass tolerance: 0.02 Da, missed cleavage: 2, and false discovery rate (FDR) <0.05. The protein abundances were normalized to the median of all proteins detected in each sample and compared between FTC and FTA using NormalyzerDE (https://normalyzerde.immunoprot.lth.se/) (13).

To estimate the expressions of genes in thyroid cancer tissues, we used the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/gds/). We analyzed the GSE15045 dataset, which comprises gene expression data from 4 FTA patients and 8 FTC patients, using the GEO2R web tool (https://www.ncbi.nlm.nih.gov/geo/geo2r/). This analysis allowed us to compare the gene expression profiles of FTA and FTC tissues.

The human FTC cell line, FTC-133, was obtained from the European Collection of Authenticated Cell Cultures (UK Health Security Agency, Porton Down, UK). FTC-133 cells were cultured in DMEM/Ham's F12 medium (FUJIFILM Wako Pure Chemical) containing 10% fetal bovine serum (FBS) (Thermo Fisher Scientific), 100 units/ml penicillin, and 100 µg/ml streptomycin (FUJIFILM Wako Pure Chemical).

Dicer-substrate short interfering RNAs (DsiRNAs) for the human RAB21 gene (109142326) and negative control DsiRNA (51-01-14-03) were purchased from Integrated DNA Technologies (Coralville, IA, USA). The sequences of DsiRNAs are shown in Table SII. DsiRNAs were transfected into FTC-133 cells at a concentration of 2 nM using the Lipofectamine RNAiMax transfection reagent (13778150; Thermo Fisher Scientific), in accordance with the manufacturer's instructions. After 6 h of incubation, the medium was changed to DMEM/Ham's F-12 medium supplemented with 10% FBS. After a 2-day culture, the transfected cells were used for western blot analysis, and cell invasion and migration assays.

DsiRNA-transfected cells were seeded at 2×10⁴ cells onto a Matrigel invasion chamber (354480; Corning, Corning, NY, USA) and incubated at 37°C under a 5% CO₂ atmosphere. After 24 h, the cells were fixed with 100% methanol and stained with crystal violet, and then, the number of invaded cells was determined. The invasion rate was represented as the ratio of the number of invaded cells for RAB21-knockdown cells compared with that for control cells.

Transfected cells were cultured in a six-well plate until they reached confluence. Cell layers were wounded using a 1 ml pipette tip and cultured for 30 h. The wound area was analyzed based on photographs taken at 0 and 30 h using ImageJ software, and the cell migration rate was calculated from the ratio of each void area at 30 h to that at 0 h, as follows:

Migration rate (%)=[1-(void area at 30 h/void area at 0 h)]x100.

The statistical significance of differences was determined using limma (14) for proteomic analysis. Unpaired Student's t-test was used for analysis of RAB21 in serum-derived EVs from FTC and FTA patients. For analysis of RAB21 mRNA expression from the GEO database, Welch's t-test was performed. For cell invasion and migration assays, data were analyzed using Mann-Whitney U test. P<0.05 was considered to indicate a statistically significant difference.

Serum EVs from patients were purified using the Tim4 molecule, which has affinity for phosphatidylserine. The purified serum EVs were confirmed by detecting EV marker proteins and measuring their particle size characteristics. The analysis of serum EVs purified from patients revealed the presence of the EV markers Alix, Flotillin-2, and CD9, which are commonly found in small EVs (Figs. 1A and S1A). Furthermore, the particle sizes were mainly distributed around 100 nm, which corresponds to the size of small EVs, and there were no substantial differences in this variable among the patients (Fig. 1B).

Proteomic analysis was performed on the EVs purified from the serum of FTC and FTA patients using an LC-MS/MS instrument to identify differentially expressed proteins. The results identified 639 proteins (Table SIII). Upon comparing the normalized protein abundances between FTC and FTA EVs, we found that 6 proteins had more than a 2-fold increase and 12 proteins had less than 0.5-fold expression in EVs from FTC patients (Fig. 2 and Table I).

Several proteins related to cancer development and progression were identified in the list of abundant proteins in FTC, including Ras-related protein Rab-21 (RAB21) (15), Keratin 13 (KRT13) (16), and Serum amyloid A1 (SAA1) (17). In contrast, the deficient proteins in FTC included antioxidative proteins such as Peroxiredoxin 5 (PRDX5), Superoxide dismutase 1 (SOD1), and Transaldolase (TALDO1). Furthermore, Four and a half LIM domains protein 1 (FHL1), which is a tumor suppressor gene in thyroid cancer (18,19), was downregulated in FTC.

RAB21, a member of the Rab family of small GTPases, has been reported to modulate cell adhesion and migration processes (15). Therefore, we focused on RAB21, and attempted to validate its upregulation in FTC serum EVs using western blot analysis, as well as in FTC tissues through in silico analyses. First, we examined the RAB21 levels in serum EVs by western blot analysis using CD9 as an EV marker and confirmed that RAB21 was more abundant in the serum EVs of FTC patients than in those of FTA patients (Figs. 3 and S1B). Second, we determined whether RAB21 was actually present in EVs, rather than being a contaminant from the blood, because proteins that are abundant in the blood inevitably contaminate EVs purified from serum. We analyzed the levels of RAB21 in blood using the Protein Atlas database (https://www.proteinatlas.org/humanproteome/blood/proteins+detected+in+ms). Compared with the other upregulated proteins, the RAB21 level was remarkably low (Fig. S2), so we considered that RAB21 was derived from EVs and was not a contaminant from the blood, and that FTC tissues might contain abundant RAB21. Third, we estimated RAB21 mRNA expression in FTC and FTA tissues using the GSE15045 dataset from the GEO public database. This suggests that RAB21 mRNA was significantly upregulated in FTC compared with the level in FTA (Fig. S3). These findings indicate that RAB21 is upregulated in FTC tissues and that its levels are increased in serum EVs from FTC patients.

Because RAB21 gene expression was increased in the tissues of FTC patients, we considered that RAB21 might be involved in the malignant phenotype of tumors. Therefore, we examined whether the knockdown (KD) of RAB21 expression in FTC-133 cells, a follicular thyroid carcinoma cell line, might affect malignant phenotypes such as invasion and migration. RAB21-KD in FTC-133 cells was performed using DsiRNA and Lipofectamine RNAiMAX, and confirmed by western blot analysis (Figs. 4A and S1C). Cell invasion was performed using a Matrigel invasion chamber for 24 h. The results showed that the invasion of RAB21-KD cells did not differ from that of the negative control (NC) (Fig. 4B). However, the cell migration rate calculated from the change in wound area during 30 h was significantly decreased in RAB21-KD cells (Fig. 4C). Taken together, these findings suggest that RAB21-KD did not affect cell invasion but reduced the migration of FTC-133 cells under these experimental conditions.